1. Field Of Invention
The present invention relates to a method and system for recognizing characters carried on a document. More specifically, the present invention relates to a novel method and system for recognizing characters, particularly those printed by hand, and particularly on a check or other financial document.
2. Background Art
Various techniques for recognizing or identifying characters on a document have been proposed in the past.
Typically these systems have used either templates or tested key regions for particular features of a character. That is, the template or overlay for each character in a set would be matched against the character to be recognized and the character would be recognized if it matched within a preset tolerance of the overlay.
Within digital processing, various techniques are known for recognizing characters through the use of a tree logic or neural network logic of asking successive questions based upon the answer to the previous question and then arriving at a conclusion as to what the character is at the end.
While these systems generally have advantages in using less storage and less processing, they have been found to be inadequate in identifying characters particularly those written by hand in a free or unconstrained manner.
Accordingly, the prior art systems for identifying characters have limitations and disadvantages which are undesirable. This is particularly true in the attempts to recognize handprint characters in unconstrained fields.
The present invention overcomes the limitations and disadvantages of the prior art systems by providing a method and system for recognizing characters.
The present invention has the advantage that it is not particularly limited to the size of the characters being recognized, the orientation of the writing or the care with which the individual features are formed. The present invention also has the advantageous effect that stray markings, either background noise or additional marks that the writer has included have a minimum effect on the recognition process. The present invention includes an optional, though desirable, method for normalizing or skeletonizing a character image which initially includes a stroke width of multiple pels into a stroke width of a single pel.
The present invention uses a test or training set to calculate historical probabilities which are used in a look up table. Depending upon the accuracy required, this training set could be expanded as necessary.
This system is also inherently flexible, in that the level of confidence to call a particular character recognized is set and reset either within the program or by operator control. That is, in a system where it is important that no character be read incorrectly and additional failure to reads are acceptable, the threshold for a successful recognition could be set very high. In another system where the penalty for an incorrect reading (a substitution) is not significant, the threshold could be set very low so that the fail to read percentage is kept very low.
This system is also flexible in that the difference in probability between the first choice and the second choice can be adjusted. That is, if the second choice is almost as likely as the first choice (within a preset range), then the character can be rejected as unread to avoid an erroneous recognition.
The foregoing and other objects and advantages of the present invention will be apparent to those skilled in the art of character recognition in view of the following description of the invention the appended claims and the accompanying drawings.